Voyage 3 Large : A newly launched multilingual universal embedding model that excels across multiple fields.

Voyage 3 Large

AI Model Development & Tools #Artificial Intelligence, Embedding Models, Multilingual, Retrieval, Quantization Standard Picks Paid

Overview :

Voyage-3-large is the latest multilingual universal embedding model introduced by Voyage AI. This model holds the top rank in 100 datasets across eight domains, including law, finance, and code, surpassing OpenAI-v3-large and Cohere-v3-English. It utilizes Matryoshka learning and quantization-aware training, supporting smaller dimensions and int8 and binary quantization, significantly reducing vector database costs with minimal impact on retrieval quality. The model also supports a context length of 32K tokens, far exceeding OpenAI (8K) and Cohere (512).

Target Users :

The target audience includes enterprises and developers who require efficient retrieval and processing of multilingual text data, such as professionals in the legal, finance, and technology sectors, as well as users with high demands on model performance and cost-effectiveness.

Total Visits： 19.8K

Top Region： US(45.24%)

Website Views ： 54.6K

Use Cases

Legal Industry: Quickly and accurately retrieve relevant legal cases and regulations to enhance legal research efficiency.

Finance Sector: Analyze financial documents and reports to quickly extract key information and assist in investment decisions.

Software Development: Understand and generate code to improve programming efficiency and code quality.

Features

Ranked first in 100 datasets across eight domains including law, finance, and code.

Supports embeddings of 2048, 1024, 512, and 256 dimensions achieved through Matryoshka learning.

Provides various embedding quantization options, including 32-bit floating-point, signed and unsigned 8-bit integers, and binary precision while minimizing quality loss.

Supports a 32K token context length, longer than OpenAI (8K) and Cohere (512).

The int8 precision version with 1024 dimensions of voyage-3-large is only 0.31% lower in retrieval quality compared to the float precision with 2048 dimensions version, yet storage costs are reduced by 8 times.

Even using 512 dimension binary embeddings, voyage-3-large still outperforms OpenAI-v3-large (3072 dimension floating-point embeddings) by 1.16%, while reducing storage costs by 200 times.

How to Use

1. Visit https://docs.voyageai.com/docs/embeddings for detailed documentation.

2. Sign up and obtain an API key.

3. Use the API key to call the voyage-3-large model with your text data.

4. Retrieve the embedding vectors returned by the model for subsequent retrieval or other tasks.

5. Select appropriate dimensions and quantization options as needed to balance performance and cost.